Video comprises the major communications and entertainment media asset, occupying more than 60% of today’s Internet traffic. Yet, video remains the least-manageable element of the big data ecosystem. This is because all state-of-the-art methods for high-level semantic description in video require either manual annotation, or compute-intensive video decoding and processing. This project aims to create a robust and performant ecosystem of machine learning algorithms to uniquely identify and describe semantic video attributes within networks and file systems (e.g., automatic semantic labeling of video segments). The tools to be used are: TensorFlow, Python, potentially a bit of
C/C++ programming, Docker containers and the Linux operating system.
It is understood that the student will not be familiar with these tools, so some summer study will be required. Team members of Dr Andreopoulos's group will be available to tutor the student on some of the practical aspects. This project is a demanding piece of work, but it is ideal for a student who is
genuinely interested to understand what deep neural networks can (and cannot) do, and the application area in video analysis is of very strong relevance to a number of industries in the UK and worldwide.